Advances in Adaptive Exploration and Cognitive Architectures

The field of artificial intelligence is moving towards developing more adaptive and autonomous systems that can efficiently explore and learn from their environments. Recent research has focused on bridging the gap between curiosity-driven exploration and competence-based control, with a emphasis on developing internal representations and world models that can facilitate rapid adaptation and effective problem-solving.

Noteworthy papers in this area include:

  • Modeling Open-World Cognition as On-Demand Synthesis of Probabilistic Models, which proposes a computational implementation of a Model Synthesis Architecture to construct bespoke mental models tailored to novel situations.
  • Assessing adaptive world models in machines with novel games, which calls for a new evaluation framework for assessing adaptive world models in AI using suites of carefully designed games with genuine novelty in the underlying game structures.
  • Behavioral Exploration: Learning to Explore via In-Context Adaptation, which proposes training agents to internalize what it means to explore and adapt in-context over the space of expert behaviors, enabling fast online adaptation and targeted exploration.

Sources

From Curiosity to Competence: How World Models Interact with the Dynamics of Exploration

Abductive Computational Systems: Creative Abduction and Future Directions

Behavioral Exploration: Learning to Explore via In-Context Adaptation

Adaptive Social Learning using Theory of Mind

Illuminating the Three Dogmas of Reinforcement Learning under Evolutionary Light

Modeling Open-World Cognition as On-Demand Synthesis of Probabilistic Models

Assessing adaptive world models in machines with novel games

Difficulty as a Proxy for Measuring Intrinsic Cognitive Load Item

Built with on top of